One of the biggest questions facing Intel as we move into 2021 is the degree to which the company will rely on third-party foundries for its leading-edge products. Over the past year or so, Intel has announced that it would tap foundry partners for a wider range of products, but left open which companies had won which business.
According to Reuters, Intel intends to manufacture DG2 — its upcoming consumer graphics card — at TSMC, on an enhanced 7nm node that hasn’t even been named yet. TSMC has sold 7nm in three flavors — N7, N7P, and N7+. N7P was the original N7 node with additional performance enhancements, while N7+ introduced EUV lithography. The introduction of EUV was a major step all on its own. TSMC may have built a new 7nm node for its own purposes; the company occasionally introduces new variants of mature nodes. 7nm is still close enough to the leading edge to plausibly be refined in some fashion.
Alternately, Intel might have paid TSMC to implement a specific version of the node that suits its own goals. Reuters claims the node will be more advanced than Samsung’s 8N, which Nvidia tapped for Ampere.
Intel CEO Bob Swan gave interviews ahead of CES 2021, in which he appears to be charting a middle ground between Intel yanking all of its current third-party semiconductor manufacturing back into its own fabs and going purely fabless. According to Swan, Intel’s goal is to maintain maximum flexibility in its approaches:
[W]e may outsource more; it means we may use more available third-party IP, it means we may make stuff for others, not just, i.e., be a foundry ourselves. And is there a scenario where we could be using somebody else’s process technology in our fabs? That’s possible. The key is, as the industry evolves, how do we leverage the innovation? Not just within our four walls, but the innovation happening in the industry as a whole, and be very flexible and adaptable to take advantage of those [innovations] along the way.
These comments aren’t too different from what Swan has said in the past, and they imply that Intel will announce a mixed strategy in which it retains its own foundries, but clarifies which products will be built at other companies on January 21. It doesn’t really matter if Intel builds DG2 at TSMC, just like it doesn’t really matter that Mobileye continues to use TSMC for its own products. When people think about Intel, they don’t think about GPUs or automotive computing. What’s really going to drive headlines and opinions about whatever strategy Intel announces on January 21 isn’t where the company builds its GPUs, IoT, automotive, or storage products. The perceived impact will turn on where future CPUs will be built and what foundry customers (if any) win which products.
According to 2020 data from IC Insights, TSMC was the second-largest foundry, with the equivalent of 2.5 million 200mm-equivalent wafer starts per month. Intel was measured at 817K wafer starts per month. While this leaves TSMC obviously dwarfing Intel, not all of TSMC’s foundries are even capable of building the kind of hardware Intel needs. Any attempt by Intel to shift its business to TSMC wholesale would also require a dramatic capacity expansion on TSMC’s part.
Look for more details on Intel’s future chip strategy to arrive on January 21.